The Missing Profits of Nations: Online Appendix

Transcription

1 The Missing Profits of Nations: Online Appendix Thomas Tørsløv (University of Copenhagen) Ludvig Wier (University of Copenhagen) Gabriel Zucman (UC Berkeley and NBER) June 5, 2018 Abstract This Appendix supplements our working paper The Missing Profits of Nations Thomas Tørsløv: Ludvig Wier: Gabriel Zucman: We thank the Danish Tax Administration for data access and numerous conversations. Zucman acknowledges financial support from the FRIPRO program of the Research Council of Norway.

2 Contents A Data on Corporate Profits Across the World 2 A.1 Main Data Sources A.2 Computation of Domestic Profits A.3 Computation of Profits of Foreign-Controlled Firms A.4 Supplementary Data on Corporate Profits B Balance of Payments Data 26 B.1 Data Sources B.2 Data on Cross-Border Flows B.3 Discrepancies in Global Direct Investment Income C Macro Statistics Corrected for Profit Shifting 30 C.1 High risk payments to tax havens C.2 Ownership of profits in tax havens C.3 Reallocating the tax haven profits D Comparisons With Previous Estimates 35 D.1 Studies Based on Financial Accounting Data D.2 Studies based on macro-data D.3 Transaction studies of transfer mispricing D.4 Benchmarking our results to previous literature E Tax Enforcement 37 E.1 Background on tax enforcement of transfer prices in practice E.2 Model of international tax enforcement F The rise of the MNE and the fall of the corporate tax 46 F.1 Data on corporate tax rates and revenue F.2 The rise of the multinational firm G List of files 48 1

3 The goal of this Appendix is to allow the reader to reproduce all the results of the paper starting from readily available public statistics. We describe each of the steps that leads from the published data to the results. The Appendix is supplemented by an Excel file containing all relevant formulas with the details of each computation and by a set of Stata files. 1 The Working Paper summarizes the main steps. The Appendix gives additional details, provides consistency and robustness checks, compares the choices made in this research with those made in other studies, lists all relevant references, and produces additional results excluded from the Working Paper for the sake of conciseness. The Appendix is structured as follows: Section A discusses the data and computation of the amount of corporate profits in each country, including the decomposition between the profits of foreign-controlled vs. local firms, and our estimates of profits artificially shifted to tax havens. Section B discusses balance of payments and trade data, and how we use these data to apportion the shifted profits to the countries where they have been made (or where the multinationals that shift profits are headquartered). Section C presents GDP, profits, capital shares, and profitability statistics for the world s main countries corrected for artificial profit shifting. Section D compares our estimates of multinationals profit shifting to previous studies. Section E provides data and information on how tax authorities attempt to enforce taxes on multinational groups (mutual agreement procedures, etc.). In this section we also present a simple theoretical model to understand the patterns observed in international tax enforcement. Section F presents data on the evolution of corporate tax rates and multinational companies share of the global economy. Section G lists the various data outputs created by this research. A Data on Corporate Profits Across the World This Section presents our database of corporate profits across the world. This database decomposes corporate profits into profits made by foreign-controlled corporations vs. local firms, and 1 Available online at: 2

4 into actual profits vs. artificially shifted profits. We describe the construction of the database step by step starting from easily accessible public statistics. All our computations are for the year 2015, the latest year for which comprehensive data was available at the time this research was conducted. The database is available in Excel format, with tables numbered A.1 to A.11. We start by presenting the data sources we use, and then discuss the construction of each of these tables in turn. A.1 Main Data Sources A.1.1 National Accounts Data The starting point to measure the corporate profits made in each country is the national accounts. National accounts data report information on value-added in each domestic sector of the economy: non-financial corporations, financial corporations, the government sector, the household sector, and non-profit institutions. In turn, value-added is decomposed into compensation of employees paid and operating surplus (i.e., profits); see Section A.2 below. By adding the operating surplus of non-financial and financial corporations, we obtain the recorded amount of profits made by domestic corporations. We use two sources of national accounts data. OECD National Accounts Data. First, we rely on the detailed OECD national accounts by sector (OECD Table 14a). 2 The OECD database includes all OECD countries and a number of large developing non-oecd countries (Brazil, China, Colombia, Costa Rica, India, Russia, and South Africa). We include all these countries in our own database. Tax Havens National Accounts Data. Second, we extend the OECD database to non- OECD tax havens by relying on the national accounts data disseminated by tax havens official statistical institutes and/or central banks. Our list of non-oecd tax havens includes Cyprus, Malta, Marshall Islands, Singapore, Hong Kong, Puerto Rico, and all the small offshore financial centers listed in Table 1 of Lane and Milesi-Ferretti (2010): Andorra, Anguilla, Antigua and Barbuda, Aruba, The Bahamas, Bahrain, Barbados, Belize, Bermuda, the British Virgin Islands, the Cayman Islands, Gibraltar, Grenada, Guernsey, the Isle of Man, Jersey, Lebanon, Liechtenstein, Macao, Mauritius, Monaco, the Netherlands Antilles, Panama, Samoa, Seychelles, St. Kitts and Nevis, St. Lucia, St. Vincent & Grenadines, Turks and Caicos, Vanuatu. Many of these tax havens publish their own national accounts, and use them whenever they exist. These national accounts are imperfect, in the sense that they typically dont attempt to es

5 timate the profits shifted inward into the offshore sector. Take the case of Bermuda, for instance. The national accounts of Bermuda provide estimates of value-added, compensation of employees paid, and operating surplus by sector of the economy. 3 They isolate an offshore corporate sector (called the international business sector) from the rest of the domestic economy. The national accounts of Bermuda report compensation of employees paid in this international business sector (namely, $1.438 billion the Bermudian dollar is equal to 1 US dollar). But they put operating surplus at 0 for this sector. In effect they do not attempt to include into Bermuda s GDP the amount of profits recorded by the foreign-controlled firms located in Bermuda, which would inflate enormously the GDP of Bermuda (hence would make even more apparent than already is the extent of inward profit shifting into the island). Therefore, to estimate the amount of profits booked (for tax reasons) in offshore tax havens, we start with the official data reported by these havens, and then make a number of step-by-step corrections described precisely in Appendix A and B below. A.1.2 Foreign Affiliates Statistics The second key data source we use is foreign affiliates statistics (FATS). While national accounts data provide information on the total amount of corporate profits made in each country, they do not show how much profits are made in foreign-controlled corporations vs. local firms. The FATS enable us to bridge this gap. Inward FATS of country A provide key economic indicators for firms operating in country A that are foreign-controlled, i.e., whose ultimate controlling institutional unit is located in a foreign country. (Outward FATS, symmetrically, provide key economic indicators for foreign affiliates of multinational companies whose ultimate controlling institutional unit is a resident of country A.) We use inward FATS to decompose the corporate profits made in each country into profits made by foreign-controlled firms vs. local firms (i.e., not foreign-controlled). A firm is foreign-controlled if a single investor or a group of associated investors acting in concert own more than 50% of ordinary shares or voting power. However, this condition is sufficient but not necessary: other criteria may also be relevant for defining foreign control, and thus other cases (multiple minority ownership, joint ventures, and qualitative assessment determining control) are sometimes used to assess control, cf. Eurostat (2012). FATS disseminated by Eurostat and the OECD. In the European Union, the first regulations mandating the compilation of FATS were introduced in In July 2007, Eurostat 3 See for

6 the EU statistical institute published a first edition of its recommendation manual for foreign affiliates statistic. A second edition was published in 2009 and a third edition in Since 2008, all EU countries generally report annual foreign affiliates statistics to Eurostat. Before 2008 a number of EU countries reported FATS statistics, but coverage was more limited (e.g., Ireland did not report data; countries reported fewer variables) and the statistics were less harmonized. Post-2008 FATS include estimates of value-added, compensation of employees paid, and gross operating surplus for foreign-controlled corporations, by sector of the economy and country of the controlling entity. However, they do not include further decompositions of gross operating surplus (into net interest paid, net dividends paid, corporate income tax paid, and depreciation). Moreover, the FATS data disseminated by Eurostat currently only cover non-financial corporations, and coverage among non-financial corporations is not always complete (see discussion in Section A.3 below). 5 The OECD also disseminates FATS data, called activities of multinational enterprises (AMNE) statistics. For EU countries, the data are identical to those disseminated by Eurostat. The OECD also includes FATS for non-eu countries: Canada, Switzerland, Turkey, and the United States. Except for the United States, these statistics are typically more limited than for EU countries (i.e., fewer variables are included). BEA Survey of Foreign Operations of U.S. Multinationals. The United States has been compiling particularly detailed data on the activities of U.S. multinational companies (and foreign multinationals operating in the United States) since the 1950s. These data are compiled by the Bureau of Economic using mandatory surveys. 6 The first modern survey of the activities of U.S. multinationals was conducted in Since 1982, a survey is conducted annually; an exhaustive (census) benchmark survey is conducted every five years. The latest benchmark survey was conducted in These data are richer than the FATS currently compiled by other OECD countries. In particular and importantly, they contain detailed decomposition of the profits made by affiliates of U.S. multinationals abroad, including foreign income taxes paid Data are available online at global-value-chains/foreign-affiliates. 6 The data are available at: companies_comprehensive_data.htm. 5

7 A.1.3 Foreign Direct Investment Statistics In a number of countries, statistics on the activities of multinational companies (i.e., FATS) are still not available. In that case we rely instead on foreign direct investment statistics to estimate the amount of profits made by foreign-controlled corporations. There are two main sources of direct investment statistics: the OECD and the IMF. The OECD sets the world standards for compiling FDI statistics through its Benchmark Definition of Foreign Direct Investment. We rely primarily on FDI statistics disseminated by the OECD. 7 These statistics follow the 4th edition of the OECD benchmark definition of foreign direct investment (BMD4) that was published in The BMD4 improved upon previous definitions along two dimensions: first it encouraged countries to compile FDI statistics separately for resident special purpose entities (SPEs), i.e., entities with no or few employees, little or no physical presence in the host economy and whose assets and liabilities represent investments in or from other countries and whose core business consists of group financing or holding activities. Second, it encouraged countries to compile inward investment positions according to the ultimate investing country to identify the country of the investor that ultimately controls the investments in their country. While BMD4 was completed in 2008, it is only since September 2014 that the OECD has been collecting FDI statistics from member countries according to the updated benchmark definition. Data for previous years followed earlier versions of the benchmark definition. We also rely on FDI statistics disseminated by the IMF when no data are available from the OECD. For OECD countries, there is generally a small discrepancy between direct investment statistics reported by the OECD and by the IMF. OECD and IMF research demonstrated that the main differences between their FDI statistics are largely due to the timing of revisions. 9 In addition to the OECD and the IMF, the UNCTAD also disseminates FDI data that are in a number of cases different than the OECD due to adjustments; we do not use UNCTAD data in this research. A.2 Computation of Domestic Profits A.2.1 From GDP to Corporate Value-Added (Table A.1) We start in Table A.1 by reporting the decomposition of GDP by sector: GDP (at factor cost, i.e. net of taxes on production) is equal to the value-added of corporations (financial plus non- 7 Available online at

8 financial corporations), plus the value-added of the government, plus the value-added of the rest of the economy (non-corporate businesses, households, and non-profit institutions serving households). The data for OECD countries and the main developing countries are taken from the OECD detailed national accounts by sector. In cols. 3 8, all taxes on production net of subsides are removed for each sector; these taxes include all taxes on production and imports (code D2 in the System of National Accounts 2008). That is, they include both taxes on products (code D21) and other taxes on production (code D29), net of the respective subsidies (codes D31 and D39). 10 For China and India we use data from their respective statistical agencies (Chinese national accounts data are from 2013, but inflated using GDP growth from 2013 to 2015). Data for South Africa and Brazil are for 2014 (with no adjustment). When no data is available, we impute the share of the corporate sector in total value-added at factor cost as the average value for similar countries (see Excel formulas in Table A.1). For non-oecd tax havens we use UN National Account data to estimate GDP at factor cost (see do-file UN National Accounts.do). To decompose GDP by production sector, whenever data is available (which is the for, e.g., Singapore, Puerto Rico) we use data from the havens official statistical agencies. If no data is available, we first try to impute current values using past values; otherwise we use average shares among non-oecd tax havens. The do-file UN National Accounts.do explains the imputation procedure in detail. A number of results are worth noting. At the global level the share of the corporate sector in total value-added is slightly higher than 60% (61.4% in 2015), but this average masks interesting heterogeneity. First, The corporate share in tax havens is particularly high: the country with the highest corporate share of domestic value-added is Ireland (80%), closely followed by Luxembourg, the Netherlands, and Switzerland. Tax havens typically have corporate shares between 70% and 80%. Second, among non-haven OECD countries, there are variations in the corporate share of total value-added. Most importantly, the U.S. share is relatively low (57.6%) due to the importance of non-corporate businesses (partnerships) and of non-profit institutions (especially in the health care sector). When one looks at the share of non-corporate businesses and non-profits in total value-added, the United States is among the countries with the highest share. Third, the share of corporate value-added in total value-added is quite similar in developing countries vs. OECD countries. The difference is that in developing countries, a relatively 10 Product taxes strictly speaking (D21) include sales taxes, value-added taxes, excise duties, import taxes and various other consumption taxes, while other production taxes (code D29) include a number of property taxes and non-social-contributions payroll taxes. 7

9 low fraction of value-added is made in the government sector and a relatively high fraction is made in non-corporate businesses (e.g., by self-employed individuals). By contrast, in OECD countries a relatively low fraction of value-added is made in the non-corporate business sector and a relatively high fraction is made in the government sector. Some of the cross-country variation in the sectoral composition of value-added also reflect a lack of harmonization in national accounts data across countries. Importantly, as pointed by Pionnier and Guidetti (2015), countries differ in the way they record the self-employed and other non-corporate businesses. Germany and Italy record certain self-employed workers in the corporate sector (and their income under corporate profits). This artificially inflates the share of the corporate sector in total value added (and affects other statistics such as effective corporate income tax rates, see below). We did not attempt to address this problem in this research and refer to Pionnier and Guidetti (2015) for a detailed discussion and plausible corrections that would make Germany and Italy more comparable to other OECD countries. A.2.2 Decomposition of Corporate Value-Added (Table A.2) Table A.2 decompose the value-added of corporations by cost component. Corporate valueadded (at factor cost, i.e., net of indirect taxes) is equal to compensation of employees paid, plus net operating surplus, plus depreciation. Compensation of employees (code D1) paid by the corporate sector includes both wages and salaries and supplements to wages and salaries (e.g., mandatory employer social contributions, employment fringe benefits such as pension contributions, etc.). Net operating surplus is equal to gross operating surplus (code B2G) minus capital depreciation (code K1). Net interest is equal to the interest paid by corporations (D41 paid) minus the interest received by corporations (D41 received). Corporate profits are computed as net operating surplus minus net interest paid, and correspond conceptually to what the corporate income tax attempts to tax (as depreciation and interest payments are typically tax deductible). We take the data from the OECD detailed national accounts by sector and countries official national accounts for non-oecd countries. We impute compensation of employees, net interest paid, and depreciation when no data is available using the mean of these variables (as a fraction of corporate value-added) for comparable countries (typically, the mean for OECD countries for OECD countries with missing data; the mean for developing countries for developing countries with missing data; the mean for non-oecd tax havens for non-oecd tax havens with missing data; see Excel formulas in Table A.2). We attribute to Bahrain and Lebanon the average labor 8

10 share of non-oecd tax havens (see Excel formulas in Table A.2); data exist for these havens but they imply implausibly low labor shares. For net interest paid by non-oecd havens (col. 4), we only have data for Singapore; we therefore proceed as follows. For Malta, Hong Kong, Cayman Islands, and Cyprus we estimate net interest paid as net FDI interest paid to the rest of the world, using balance of payments data. For other non-oecd tax havens, net interest paid is imputed using the weighted average share of net interest paid to corporate value-added in these five non-oecd tax havens. In cols. 7 9 we compute factor shares (the share of labor and the share of capital in total corporate value-added), both gross of capital depreciation and net of capital depreciation. We also report in col two of our key ratios of interest in this research: the ratio of net interest paid to net operating surplus (col. 10), and the ratio of taxable corporate profits (defined as operating surplus net of capital depreciation and of net interest payments) to compensation of employees. A number of results are worth noting. First, we observe very high capital shares in tax havens compared to all other countries, and accordingly high profits to compensation ratios. However tax havens are not the only countries that have capital share and profits/compensation ratios; a number of developing countries also do (Mexico, India, Turkey, etc.), as well as a number of resource-rich countries (Chile, Norway, Russia, South Africa). Developed, high-tax countries tend to have the lowest capital shares (France, Canada, Finland, Belgium, United States, Austria, Spain, Austria, etc.). The patterns are similar when looking at the corporate capital share within non-financial corporations only, i.e., excluding financial corporations (Table A.2b). The high capital share of corporate value-added in developing countries means that a large fraction of global profits come from developing countries today. When one ranks countries by the size of their corporate profits (in US$ using market exchange rates), then the number one country in the world is not the United States (as when ranking countries by GDP), but China. India is 6th, Mexico 7th, Russia 8th, and Brazil 9th. By contrast, France (which is the 7th largest country by size of GDP) is only 13th by corporate profits. Second, there is substantial variation in the ratio of net interest paid to operating surplus. Net interest paid by corporations in high-tax countries is typically positive while net interest paid by corporations in tax havens is typically negative. Part of this reflects the fact that tax havens host a large financial industry, which typically receives positive net interest. In Appendix Table A2.b, we provide a decomposition of the value-added of the non-financial corporate sector. 9

11 We see here than non-financial corporations in all countries have typically positive net interest payments, with again substantial variation across countries. Net interest paid by non-financial corporations are particularly high in Canada, France, and the United States potentially reflecting a greater use of interest payments for tax avoidance purposes in these countries. They are also very high in Luxembourg, potentially reflecting income payments of hybrid securities (i.e., securities treated as bonds for tax purposes in Luxembourg and equity for tax/regulatory purposes in other countries; these securities are commonly used to avoid corporate taxes, see Johannesen, 2014). A.2.3 Distribution of Corporate Profits (Table A.3) Table A.3 decomposes corporate profits into net dividends paid, corporate income tax paid, and retained earnings. Corporate profits are the profits reported in col. 5 of Table A.2, i.e., operating surplus net of capital depreciation and of net interest payments typically what the corporate tax attempts to tax. Dividends include the distributed income of corporations (code D42 in the SNA) plus investment income disbursement (code D44), which includes investment income attributable to insurance policy holders, payable on pension entitlements, and attributable to collective investment funds. Retained earnings (col. 5) are computed as a residual, hence include net rents (code D45, usually zero or negligible) and net business transfers (such as fees paid to the government, fines, donations, etc.; usually small) in addition to pure retained earnings. In the national accounts, corporate income tax payments include the profits of central banks (which by convention are treated as if they were 100% taxed by governments). Because these profits have increased after the financial crisis of , they can bias computations of effective corporate tax rates based on national accounts data. Therefore in column 7, we report the actual corporate tax revenue received by governments of each country as reported by the OECD in its tax revenue statistics. 11. The difference (col. 8) can be interpreted as the profits of central banks (and other measurement and conceptual differences between the national accounts and the OECD revenue statistics). A number of results are worth noting. First, as shown in col. 9, the global average corporate income tax rate (defined as corporate income taxes paid over corporate profits recorded in the national accounts, after net interest payments) is a bit below 20% (19.3%). Unsurprisingly, effective corporate tax rates are particularly low in the main tax havens, Ireland, Luxembourg,

12 and the Netherlands; they are also low in most Eastern European countries (Poland, Latvia, Hungary, Estonia, etc.) that have low statutory rates. Germany also shows up with a low corporate tax rate of 11%. One likely explanation is that German figures for corporate valueadded and profits are distorted by the inclusion of the self-employed in the corporate sector (Pionnier and Guidetti, 2015). 12 Because the self-employed do not pay corporate taxes, the inclusion of the self-employed in the corporate sector (and of their income under corporate profits) biases effective tax rates downwards. Moreover, the corporate income tax payments recorded in the OECD revenue statistics (which we use to compute the effective corporate tax rate) for Germany are markedly lower than the corporate income tax payments recorded in the national accounts; using the latter, the German effective corporate income tax rate would rise to 14.4%. It is unclear why there is such a large gap between corporate tax revenue in the national accounts vs. the revenue statistics for Germany. Last, it is possible that there is sizable tax avoidance by German firms (and/or that many German corporations do not have to pay the corporate income tax), which could explain why the effective rate is substantially below the statutory rate of about 30%. Effective corporate income tax rates are high in countries that have high statutory rates, mainly large developed countries: the United States, Japan, France, Scandinavian countries. Interestingly, effective corporate tax rates are also high in Australia and Canada, where the corporate tax is integrated with the personal income tax, which in principle reduces the incentives for corporate tax avoidance. A few developing countries have relatively high effective tax rates (Colombia, South Africa), but most have low or very low rates (India, India, Mexico, Russia). Turning to patterns in retained earnings, column 10 shows that countries vary a lot in the fraction of after-tax corporate profits that are distributed vs. retained. For the world as a whole, slightly more than half of post-tax corporate profits are retained. There has been a sharp increase in corporate retained earnings in recent years (Chen, Karabarbounis and Neiman, 2017). Retained earnings are higher in developing countries (maybe due to more binding credit constraints) than in OECD countries (around 70% vs. around 40%). Within OECD countries, there are extreme variations. In the Netherlands almost 100% of profits are retained earnings, maybe reflecting profit shifting and tax avoidance (e.g., by U.S. multinational companies, which until 2018 had incentives to retain profits offshore). In Luxembourg by contrast, retained earnings are negative, which could be due to several things. By construction, for the mutual fund industry (which is particularly large in Luxembourg) all profits are distributed (under code 12 As pointed by Pionnier and Guidetti (2015), the same problem occurs in Italy, which also shows up with a relatively modest effective corporate tax rate of 18%, way below the statutory rate of 31.4% in force in

13 D44 in the national accounts). Moreover, for non-financial corporations, figures for Luxembourg may be distorted by the use of hybrid securities, as discussed below. A.3 Computation of Profits of Foreign-Controlled Firms A.3.1 Corporate Value-Added: Local vs. Foreign-Controlled Firms (Table A.4) Table A.4 decomposes corporate value-added into the value-added of foreign-controlled firms and the value-added of other firms (not foreign-controlled). Following internationally-agreed guidelines, foreign-controlled firms include all firms where foreign investors own more than 50% of shares with voting rights. However this condition is sufficient but not necessary: there are some other ways firms can be foreign controlled (see Eurostat, 2012). The key data source is the inward FATS statistics disseminated by the OECD, the Bureau of Economic Analysis, and Eurostat (see section A.1 above). Specifically, for European Union countries, we use the FATS disseminated by Eurostat, for the United States we use the FATS disseminated by the Bureau of Economic Analysis (majority-owned affiliates of U.S. multinationals, see Section A.1 above), and for other OECD countries we use the OECD FATS when data exist. A few remarks are in order about foreign affiliates statistics. First, the FATS disseminated by the OECD and Eurostat currently only cover non-financial corporations (except for the United States). This means in particular that they exclude financial holding companies, including special purpose entities (SPEs). Second, coverage among non-financial corporations is not always complete. In Table A.9, we reconcile foreign affiliates statistics with national accounts data. In most countries, the value-added of non-financial corporations recorded in the FATS (for both foreign-controlled and local firms) adds up to around 90%-95% of the value-added of non-financial corporations recorded in the national accounts. In some countries, however, the coverage ratio is lower, e.g., France (83%), Spain (79%), and a number of Eastern European countries. This could be due to the fact that some countries collect data from non-financial corporations based on a sample rather than an exhaustive census. Surveys face non-response problems, especially when answering the survey is not made compulsory by law. Therefore, to estimate the total amount of employee compensation and profits made by foreign-owned corporations in Table A.4, we proceed as follows. We compute what fraction of employee compensation and profits are made by foreign-owned non-financial companies in the FATS, and we apply this ratio to the total employee compensation and profits of domestic firms 12

14 (financial and non-financial) as recorded in the national accounts. 13 When compensation of employees is not reported in the inward FATS (or where no inward FATS are disseminated, which is the case for most developing countries and non-oecd tax havens), we impute it by assuming that non-u.s. affiliates have the same profitability as U.S. affiliates. Specially, we apply the profits / compensation ratio of U.S. affiliates (as reported in the outward FATS of the United States) to the total amount of profits in the foreign-controlled sector, as estimated in Table A.5 below using balance of payments statistics. 14 When no data is available, we impute the share of compensation which is paid in the foreign-controlled sector by using the weighted average share for similar countries, see Excel formulas in Table A.4. A number of results are worth noting. First as shown by col. 8, the value-added of foreigncontrolled firms accounts for 12% of global corporate value-added: 15% in OECD countries and 9% in developing countries. Among OECD countries, tax havens (Luxembourg, Ireland, the Netherlands) and Eastern European countries (most prominently Hungary, Slovakia and the Czech Republic) appear to be largely foreign-owned, with more than 40% of all corporate value-added made in foreign-controlled corporations (and as much as 65% in Luxembourg). In tax havens, these high ratios are driven by the fact that an even fraction of total domestic corporate surplus is made in foreign-controlled corporations (col. 10): as much as 75% 80% of all profits made in Ireland and Luxembourg are made in foreign-controlled firms. In Eastern European countries, a high fraction of both compensation of employees (col. 9) and profits is made in such firms. At the opposite end of the spectrum, large economies whether developed or developing tend to have low ratios of foreign ownership: China, Turkey, Japan, India, the United States and Korea all have around 10% or less of their corporate value-added made in foreign-controlled firms. In Table A.4 we also report estimates of the value-added of foreign-controlled firms in non- OECD tax havens. However, it is important to keep in mind that these figures severely underestimate the importance of foreign-controlled businesses, for two reasons. First small tax havens typically do not record the profits made by offshore firms accurately (or even not at all). Take the case of Bermuda, already mentioned. The national accounts of Bermuda provide estimates of value-added, compensation of employees paid, and operating surplus by sector of the economy. 15 They isolate an offshore corporate sector (called the international business sector) 13 For Luxembourg, we assume that 100% of the value-added in the financial sector (which is not reported in the FATS) is in foreign-controlled firms. 14 For Brazil and Russia we use the U.S. affiliate profitability of 2014 instead of 2015, as most U.S. affiliates are in the oil sector and 2015 profitability ratios are affected by the collapse in oil prices in See for

15 from the rest of the domestic economy. The national accounts of Bermuda report compensation of employees paid in this international business sector (namely, $1.438 billion the Bermudian dollar is equal to 1 US dollar). But they put operating surplus at 0 for this sector. In effect they do not attempt to include into Bermuda s GDP the amount of profits recorded by the foreign-controlled firms located in Bermuda, which would inflate enormously the GDP of Bermuda (hence would make even more apparent than already is the extent of inward profit shifting into the island). Second, the data reported in Table A.4 exclude special purpose entities (and the profits shifted into such entities). We correct for these two issues in Table A.6 below. The main limit of current foreign affiliates statistics is that they do not decompose gross operating surplus into net dividends, net interest, corporate tax paid, retained earnings, and depreciation. For some havens, most spectacularly Luxembourg, a lot of profit shifting is done through interest payments. Operating surplus is not affected by intra-group interest payments, so operating surplus alone is not informative of the full scale of profit shifting. To bridge this gap, we need to decompose the operating surplus of foreign-controlled corporations, a task we now turn to. A.3.2 Operating Surplus of Foreign-Controlled Companies (Table A.5) Table A.5 decomposes the gross operating surplus of foreign-controlled companies (excluding SPEs) into net interest paid, net dividends paid, retained earnings, corporate income tax paid, and depreciation. Because foreign affiliates statistics do not currently provide such details, these components must be estimated from other sources. To do so, we use balance of payments statistics on direct investment (DI) income. Conversely, in Table A.5 we also construct estimates of the gross operating surplus of foreign-controlled companies for the countries that have no foreign affiliates statistics (mostly non-oecd tax havens) by using balance of payments statistics on direct investment income. Consistency between FATS and DI statistics Direct investment data capture the crossborder interest, dividends, and retained earnings flows of firms who are more than 10% owned by foreign investors. These flows are apportioned proportionally to what fraction of equity is foreign-owned. This is not the same definition as the definition of foreign-controlled used in FATS statistics (which typically cover firms that are more than 50% owned by foreign investors, with no apportionment by equity ownership). Hence our decomposition of the operating surplus of foreign-controlled corporations has some margin of error. But this margin of error is usually relatively small, because in practice there is sizable overlap between foreign-controlled firms and 14

16 DI firms. Therefore as a baseline we use the DI balance of payments data with no adjustment, and we conduct a number of sensitivity tests and checks to make sure that our imputations deliver sensible results; we also always make sure that all adding up accounting constraints are respected. The only systematic correction we make to the DI data is to remove the flows of special purpose entities, since these SPEs (and financial corporations more broadly) are not included in foreign affiliates statistics. Following the implementation of the OECD 4th benchmark definition of direct investment, compiling countries have been encouraged to publish direct investment statistics separately for SPEs and non-spes (i.e., operating units). When no information in the DI flows of SPEs exists, we assume these flows are zero. We now describe how we decompose the operating surplus of foreign-controlled firms component by component. Net interest paid. We estimate the amount of net interest paid by foreign-controlled companies (col. 2) as the amount of net interest paid on inward direct investment recorded in the balance of payments (excluding SPEs). This is equal to the difference between (i) interest paid to foreign parents and fellow enterprises with a foreign ultimate controlling parent (interest paid, col. 3), and (ii) interest received from foreign parents and fellow enterprises with a foreign ultimate controlling parent (interest received, col. 4). All of these interest flows are reported in balance of payments statistics that follow the 6th edition of the IMF balance of payments manual, and are reported in Tables B.3, B.4, and B.5 in Appendix B below. We borrow the numbers from these Tables, further described below. Note that by construction, our measure of the net interest paid by foreign-controlled companies only takes into account intra-group interest flows. That is, it disregards any interest payments made to/received from firms outside of the multinational group (such as interest payments on money borrowed from unaffiliated banks). These non intra-group interest flows are likely to be small, as most of the financing of the affiliates of foreign multinationals is typically intra-group. Disregarding these flows has no material impact on our estimates. Net dividends paid. We estimate the amount of net dividends paid by foreign-controlled companies (col. 5) as the amount of net dividends paid on inward direct investment recorded in the balance of payments (excluding SPEs). This is equal to the difference between (i) dividends paid to foreign parents and fellow enterprises with a foreign ultimate controlling parent (dividends paid, col. 6), and (ii) dividends received from foreign parents (typically 0) and from 15

17 fellow enterprises with a foreign ultimate controlling parent (dividends received, col. 7). All of these dividends are in principle reported in balance of payments statistics that follow the 6th edition of the IMF balance of payments manual (see Tables B.3, B.4, and B.5 in Appendix B below). Retained earnings. The retained earnings of foreign-controlled corporations are equal to the amount of reinvested earnings on inward foreign direct investment recorded in the balance of payments, excluding SPEs (as reported in Table B.4 and B.5 below). Corporate income tax. One limitation of current foreign affiliates statistics is that they do not report the amount of corporate income taxes paid by foreign-controlled corporations. To our knowledge, only one country reports such information in its own outward FATS: the United States, in the BEA survey of the foreign operations of U.S. multinationals (see section A.1 above for a presentation of this dataset). Therefore, in col. 9 we estimate the amount of taxes paid by foreign-controlled corporations by applying the effective tax rate faced by all domestic firms (from Table A.3, col. 9) and for tax havens and a few other countries we use instead the effective rate faced by U.S. affiliates (reported in col. 16); see Excel formulas in Table A.5. We compute the effective tax rate faced by U.S. affiliates in foreign countries using the BEA survey of the the foreign operations of U.S. multinationals. For both domestic firms and U.S. affiliates, the effective rate is similarly computed as the ratio of income taxes paid to pre-tax corporate profits (i.e., net of depreciation operating surplus minus net interest paid). 16 Depreciation. We compute depreciation in foreign-controlled corporations as a residual, i.e., as gross operating surplus, minus net interest paid, net dividend paid, corporate income taxes paid, and retained earnings. We checked that the implied depreciation is reasonable, i.e., that the ratio of depreciation to gross operating surplus for foreign-controlled corporations is similar to the ratio recorded for all corporations and for U.S. affiliates (see cols ). This is the case in all countries (suggesting that our imputation of net interest, dividends, retained earnings, and taxes for foreign-controlled corporations delivers reliable results), except in the case Luxembourg. The discrepancy for Luxembourg probably owes to differences in scope 16 In the United Sates, the BEA provides a decomposition of the operating surplus of the foreign affiliates of U.S. multinationals which we report in Table A.10 (for 2015) and Table A.10b (for 2014). It also reports a decomposition of the net income of foreign affiliates of U.S. multinationals which we report in Table A.11 (for 2015) and Table A.11b (for 204). Pre-tax corporate profits is equal to what BEA calls profit-type return (Table A.10) and can equivalently be computed as net income plus foreign income taxes paid minus income from equity investments minus capital gains (Table A.11). 16

18 between FATS and DI statistics. Therefore for Luxembourg, we simply set depreciation rates in the foreign-controlled sector equal to the depreciation rate observed for the entire corporate sector; we assume that the effective corporate tax rate is the same as in the entire corporate sector, we assume that net dividends and retained earnings are accurately estimated using DI statistics, and we compute net interest paid as a residual; see formulas in Table A.5. In Table A.5, we also estimate the gross operating surplus of foreign-controlled corporations for the countries where no (or insufficiently detailed) FATS exist, namely Australia, Canada, Chile, Iceland, Israel, Japan, Korea, Mexico, New Zealand, Switzerland, Turkey, non-oecd developing countries, and non-oecd tax havens. For these countries, we estimate net interest paid, net dividend paid, and retained earnings from balance of payments statistics just as described above. We estimate corporate income taxes by applying the average corporate tax rate observed for the overall domestic corporate sector (or for affiliates of U.S. multinationals). 17 We compute depreciation by applying the rate of depreciation observed for the overall domestic corporate sector (or for affiliates of U.S. multinationals); see Excel formulas in Table A A.3.3 Corrected Corporate Profits (Table A.6) In Table A.6, we correct the estimates of corporate profits presented so far to account for the profits that go unrecorded in the national accounts and/or international investment data that we relied on until now. The are two types of profits that go unrecorded. First, non-oecd tax havens typically do not measure the profits made in the offshore sector (cf. the case of Bermuda already discussed, where the operating surplus in the international business sector is set to zero in the official national accounts). Second, even in countries that record profits made in the foreign-controlled sector, these profits are sometimes under-counted. This is the case for the E.U. tax havens, which pay less direct investment income than what partner countries say they receive from them. E.U. havens, in particular, substantially under-estimate the amount of profits made by affiliates of U.S. multinationals. In Table A.6. we correct for these two problems. The general principle guiding our correction is the following: our goal is to ensure that at the global level, the total profits made by affiliates as reported by the countries where affiliates are located add up to the total profits made by affiliates as reported by the countries where the parents are located. That is, our target is to 17 For non-oecd tax havens, we apply the statutory rates (usually in the range of 0% to 5%). 18 Using available balance of payments data, Malta shows up as having much more profits in foreign-owned corporations than in the total economy (probably due to inconsistencies between balance of payments statistics and national accounts statistics); therefore for Malta we set the gross operating surplus of foreign-controlled corporations to 0. 17

19 ensure that the global balance on direct investment income sums up to zero, which seems a reasonable requirement. As we show in detail in Appendix B below, in the available data this is not the case: at the global level, according to the IMF balance of payments statistics, there is each year more direct investment received than paid (see shown in Table B.9). That is, the world runs a direct investment income surplus. This surplus is large and growing: in 2015 it reached about $200 billion. This problem has two reasons. First a number of countries, most importantly Caribbean tax havens (e.g., the British Virgin Islands), do not publish balance of payments statistics; or when they do (e.g., Bermuda, the Cayman Islands, the Bahamas), they only report incomplete data (with no or very incomplete direct investment income data). Because these territories are used by multinational companies to shift profits, they are likely to have a negative direct investment balance (i.e., profits are being booked in these territories that accrue to foreign parents), which contributes to explaining why the world as a whole seems to run a direct investment income surplus. Second, there is measurement error in DI income statistics and inconsistencies in the definitions used across countries. For instance some countries may miss some affiliates of foreign multinationals (e.g., shell companies); the coverage of real estate is imperfect (in principle, according to BPM6 guidelines, cross-border real estate holdings should be recorded as direct investment, but not all countries collect the data necessary to estimate the related flows and positions accurately); not all countries apply the OECD 4th benchmark definition of direct investment consistently (e.g., some countries may apply a different ownership threshold than the 10% mandated by the OECD). Therefore we proceed in two steps. Step 1. First, we correct the data of the main E.U. tax havens: Belgium, Ireland, Luxembourg, and the Netherlands. We base our correction on a systematic investigation of the discrepancies in bilateral direct investment income data when both the investor and the host countries report bilateral DI income statistics (Table B.11 below). This investigation reveals that the European Union tax havens under-estimate the DI income they pay by $107 billion in Almost all of this gap owes to a large gap ($95 billion) between what these havens report paying to the United States and what the United States declares receiving from these havens in fact, with other partners, there is almost no discrepancy. We add these unrecorded profits to the amount of profits recorded by Belgium, Ireland, Luxembourg, and the Netherlands in their foreigncontrolled corporations (Table A.6, col. 1 and 7). 19 This adds more than $100 billion in profits 19 Direct investment income is net of corporate taxes; because we are interested in estimating pre-tax corporate profits, we upgrade the missing DI income flows using a low corporate tax rate, see Excel formulas in Table A.6. 18